Scaling

Scaling is two problems wearing one name. The architectural problem is designing a system whose parts can grow independently, so that adding capacity where it is needed does not mean adding it everywhere. The operational problem is deciding when to add it and on what signal.

Scaling concepts and services covers the first: loose coupling, scaling out versus scaling up, and — the decision that most often goes wrong — choosing the metric a scaling policy should track. Event-driven architecture on AWS takes the asynchronous form of the same idea further, into events, routing, and the choice between orchestrating a workflow and choreographing one.

Auto Scaling on AWS covers the second: which of the three services named “Auto Scaling” applies to which resources, the four modes an Auto Scaling group can be operating in, and why launch templates have replaced launch configurations outright.

Compute Optimizer closes the loop by telling you whether the capacity you settled on was the right size in the first place.